基于频谱分析和自组织神经网络的火焰燃烧诊断研究

Research on Flame Detection and Combustion Diagnosis Based on Spectrum Analysis and with Self Organized Neural Networks

通过对实验室煤粉燃烧炉实验数据的采集与分析,介绍了一种基于FFT变换处理与自组织神经网络状态识别相结合的燃烧诊断方法。首先,通过光电传感器获得一系列以一定的频率、在某个均值左右上下波动的火焰强度值。然后,利用FFT程序将获得的时域信号转换成频域上的功率谱信号。因为在稳定和不稳定的燃烧状态下,转换后得到的低频分量有明显的区别,所以把每个功率谱中前30个低频分量取出,将其作为神经网络的训练输入。通过自组织训练,神经网络将得到对应于稳定和不稳定燃烧状态火焰信号的不同输出区域。经过验证,这种方法能非常有效地识别燃烧火焰状态的稳定与否,在信号采样频率的选择,神经网络算法的改进等方面作了有意义的探索。

A combustion diagnosis method for the diagnosis of combustion, based on combined FFT transformation and mode identification with self-organised neural networks is being presented, which has been used after the acquisition of test date from the lab's coal firing furnace.First,a series of flame intensity data ,acquired via an optoelectronic sensor and which flectuate with a certain frequency around some mean value ,are converted from the time plane to the frequency plane by a FFTprogram.Because remarkable difference exists between converted low frequency components of stable and those of unstable combustion,the first lowest 30 ones of each flame's power spectrum are picked out to be used as the neural network's input signals for training porpose.By self-organised training the network builds up distinct output maps corresponding to stable and unstable flame state signals.Verification shows that this method can very efficiently distinguish unstable from stable combustion conditions.Potant probing work has been done concerning choice of frequency sampling and improvement of neural network arithmetic.Figs 4,tables 3 and refs 12.